Yapay sinir ağları kullanılarak Türkiye gün öncesi piyasası elektrik fiyat tahmini
Turkish day ahead market electricity clearing price forecasting using artificial neural network
- Tez No: 515145
- Danışmanlar: DOÇ. DR. GÜLGÜN KAYAKUTLU
- Tez Türü: Yüksek Lisans
- Konular: Enerji, Energy
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2017
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Enerji Enstitüsü
- Ana Bilim Dalı: Enerji Bilim ve Teknoloji Ana Bilim Dalı
- Bilim Dalı: Enerji Bilim ve Teknoloji Bilim Dalı
- Sayfa Sayısı: 123
Özet
Dünya genelinde ülke ekonomilerine olan ciddi etkileri dikkate alındığında uygun enerji politikaları ile enerji sektörünün planlı gelişmesinin önemi bilinmektedir. 2001 yılında yürürlüğe giren 4628 sayılı Elektrik Piyasası Kanunu ile hız kazanmaya başlayan Türkiye Elektrik Piyasası serbestleşme süreci, 2013 yılında yürürlüğe giren 6446 sayılı Elektrik Piyasası Kanunu kapsamında devam etmektedir. 6446 sayılı Kanununun amacı“elektriğin yeterli, kaliteli, sürekli, düşük maliyetli ve çevreyle uyumlu bir şekilde tüketicilerin kullanımına sunulması için, rekabet ortamında özel hukuk hükümlerine göre faaliyet gösteren, mali açıdan güçlü, istikrarlı ve şeffaf bir elektrik enerjisi piyasasının oluşturulması ve bu piyasada bağımsız bir düzenleme ve denetimin yapılmasının sağlanması”olarak ifade edilmiştir. Türkiye Elektrik Piyasası kamu santrallerinin baskın olduğu durumdan günümüze kadar gerek arz ve talep bileşenleri açısından, gerekse regülasyonlar ve modelleme açısından piyasa çeşitli dönüşüm süreçlerinden geçmiştir. Tüm bu süreçlerde Elektrik Piyasasında saatlik oluşarak enerjiyi fiyatlayan rakamlar sektördeki paydaşların ve yatırımcıların yakından takip ettiği ve çeşitli yöntemlerle tahmin etmeye çalıştığı en belirleyici gösterge niteliği taşımaktadır. Çalışmasda veri olarak kullanılan geçmiş tarihli fiyatlar piyasa işletme görevini yürüten Enerji Piyasaları İşletme Anonim Şirketi (EPİAŞ) tarafından şeffaflık platformunda kamuya açık bir şekilde paylaşılmakta olan Piyasa Takas Fiyatları (PTF)'dır. Piyasadaki saatlik elektrik arzı ve talebinin takası sonucu oluşan Piyasa Takas Fiyatları (PTF) elektriğin diğer ticari platformlardaki işlemleri için referans unsuru olduğu için literatürde de Referans Fiyat kavramına karşılık gelmektedir. Bu tez çalışması kapsamında tek gizli katmanında farklı nöron sayıları içeren Levenberg-Marquardt geriye yayılmalı öğrenen Yapay Sinir Ağları algoritması kullanılarak, Türkiye Elektrik Piyasası kısa dönemli Referans Fiyat tahmini çalışması gerçekleştirilmiştir. Türkiye'de bir çok fiyat tahmini çalışması yapılmıştır ancak, bu çalışmada modelde ele alınan girdilerin etkileri ve hangi girdiler ele alındığında en yakın sonuçların elde edildiği çalışılmıştır. Girdiler, farklı zamana bağlı fiyat serileri kadar, yerel pazarın özelliklerini de taşımak zorundadır. Türkiye elektrik piyasasında mevcut ve olası elektrik piyasası aktörlerinin faydalanabilecekleri düşünülerek yapılan bu tez çalışmasında, zamanın koşullarına göre iyileştirilen/geliştirilen YSA modelleri ile gerçekleşecek PTF'lere sinyal verecek düzeyde tahminler yapılabileceği gösterilmektedir.
Özet (Çeviri)
Electricity is a different type of commodity regarding that it cannot be stored in considerable quantities yet. Thus bringing supply together with the needed demand in a grid environment is the main system requirement. Supply side consisting of energy generation companies, show better results in a deregulated market structure. Liberalization of the energy markets has started during the end of eighties however, it is still an ongoing process in some countries like Turkey. Turkish Electricity Market started its liberalization process in 2001 by the Law no. 4628 and continues the process under the Law no. 6446 released in 2013. The main mission of deregulation is to enable the generation activities more effective and minimize electricity costs. With the less regulated energy market structure, the market becomes more competitive. In a day-ahead market, participants submit their bids for selling and buying electricity for the next 24 hours and market is cleared at the prices where demand meets supply for every single hour ahead for 24 hours. The clearing prices achieved for the day-ahead market are accepted as reference prices and gives valuation signals for the upcoming decisions through the other trading mediums such as intra-day, futures and forward markets. It is a known fact that in a more competitive market environment price forecasting has become a critical factor for the decision-makers amongst the market participants. Market participants have to forecast day-ahead prices to decide on their bidding strategies in the right way in order to maximize their profits in the spot markets. Trading companies also require price forecasts in different time horizons in order to negotiate and trade bilateral contracts. Risk managers need price forecasts to assess market risks and decide on hedging strategies timely. Investors require long term price forecasts to take feasible decisions of investing in energy sector. Therefore, well-performing forecast configurations for day-ahead electricity prices have a great importance from all perspectives and matter that much for electricity market participants of all kind. By focusing on this issue, this study is designed with the motivation of generating the most performant ANN model serving for day-ahead electricity price forecasting for the Turkish power market. Non-linearity and high volatility features of the electricity prices make forecasting a very complex task. In recent years, artificial intelligent methods are mostly used for price forecasting instead of traditional price forecasting methods. Among modern techniques such as ANN and Fuzzy logic, ANN method is a powerful tool for forecasting. The reason is that following a training process with an optimum input set, ANN method is capable of learning complicated input-output relationships. The performance of an ANN model is extensively based on the selected input set. Therefore the goal of input selection in this study is to configure an optimum set of inputs. Putting optimal inputs would result a smaller ANN model. Thus, the network would have a higher convergence speed. There are too many factors that have an impact on electricity prices. Some of those factors such as the strategical bidding/offering behaviors of the buyers/sellers are unknown and unpredictable parameters. Besides those unknown parameters, there are other parameters which are affecting the electricity prices in some extent. Some of those factors have a greater role than the others. To be more practical, only those factors which are relatively more significant are considered in this study. At first the historic prices are selected to involve in the input set. Historic prices gives an effective insight to the model for future predictions. Previous day's realized market prices and realized market prices seven day ago at the same day type are the historic prices included in the input data set. It is a known fact that electricity prices are affected by electricity demand amount which is also defined as system load. Electricity demand is depended on the weather conditions such as temperature and the extent of both industrial and commercial business operations. Generation from renewable energies especially wind energy resources in Turkish case, are compansating the demand amount in the grid. Thus, while preparing the demand inputs for the model, the residual demand after the wind energy generation amount is deducted is considered. Parameters which are related to the price forecasting can be categorized into day type. One of the day type parameters is the day of the week. Thus the model could also distinguish the weekdays from weekends. The other day type parameter is the indicator if the day is in the Ramadan month or not. The demand profile in Ramadan month differs from regular days. The other day type parameter is the indicator if the day is in the Eid period or not. The demand decreases in Eid days compared to the regular days. Since conventional energy resources have a cost-based selling strategy, the cost of the resource is directly affecting the electricity prices. For this reason, another input parameter is selected as the cost of natural gas. When the demand rises up, the market price is cleared with the offered prices of natural gas based generation power plants. Thus the cost based offer prices constitute the marginal prices in the merit order according to the efficiency of the power plant technology. In the natural gas price input set there exists a %10 discount for natural gas price starting from 1st October 2016. Hence the network is trained by taking the discount into consideration. Another parameter is selected as hourly temperature meterings. In cases when the temperature rises and falls in other words out of the mild season, electricity demand increases correspondingly. Therefore the electricity prices rise consequently. For teaching the corelation between temperature and prices, temperature meterings are included into the input set. The input set initial date is starting from 1st June 2016. The reason behind is the displacement of the algorithm stating electrcity prices at that date. For eliminating algorithm effect, the network is not trained with the inputs before 1st June 2016. Since June 2016 this optimization model originating by local developers is being used in day-ahead market operated by Energy Exchange Istanbul (EXIST). The ANN model is constructed as a single hidden layer model that has nonlinear log-sigmoid (logsig) transfer function. The selected training function for the studied ANN model is Levenberg-Marquard back propagation which is faster compared to conventional back propagation methods. That function updates weights and bias values. Throughout training, the weights and biases of the model are calibrated due to minimise the network performance function. In order to maximize the performance of the model, the hidden layer is tested with five different neuron cases; which have 7, 10, 20, 22 and 25 neurons. The time series of prices are tested for different time lags. Furthermore, local electricity market influencers are altered as to taking the total demand or net wind effect. Best performance result with 7.9% of MAPE is achieved when the ANN model has 20 neurons in its hidden layer with the most performant input set indicated as: Price from the same hour of the previous day Price from the same hour and same day of the previous week Day of the week (1-7) Day of the Ramadan month indicator (0 or 1) Day of the Eid-holiday indicator (0 or 1) Dry bulb temperature Natural gas price Net system load which is defined by considering wind generation effect on total energy demand It is noted that the wind energy generation in the input set decreased the MAPE by 1.01%. Whereas, time lags for 14, 21 and 28 days had negative effects on prediction. Besides, if the time series of historic prices is extended more, i.e., including prices of 14, 21 and 28 days ahead prices to the input set respectively, it is noted that the forecast performance is effected negatively with increasing MAPE indicators. Additionally, weight and bias matrices between the input layer and the hidden layer, as well as the weights between the hidden layer and output layer are investigated in this study. It is noted that the most influencial input parameters are day type indicators consisting of Eid-holiday, Ramadan and day of the week. Moreover, natural gas cost is also a notable influencer due to the input weights. There are quite a few studies on electricity price forecasting in Turkey. However, this study will be supporting the forecasting people in combining the time series of prices and other local market influencers as input. The proposed model is designed to perform better by ranking the input weights and changing the hidden layer design.
Benzer Tezler
- Türkiye elektrik piyasası kısa dönemli referans fiyat tahmini
Turkish electricity market short term market clearing price forecasting
SERCAN YILDIZ
Yüksek Lisans
Türkçe
2015
Ekonometriİstanbul Teknik ÜniversitesiEnerji Bilim ve Teknoloji Ana Bilim Dalı
PROF. DR. SERMİN ONAYGİL
- Essays on electricity price modeling and forecasting
Elektrik fiyatlarının modellenmesi ve tahmini üzerine makaleler
UMUT UĞURLU
Doktora
İngilizce
2019
İşletmeİstanbul Teknik Üniversitesiİşletme Mühendisliği Ana Bilim Dalı
PROF. DR. OKTAY TAŞ
- Türkiye elektrik dengeleme ve uzlaştırma piyasasındaki sistem gün öncesi fiyatının yapay sinir ağları kullanılarak modellenmesi
Modeling the day ahead price at electricity balancing and settlement market of Turkey by using artificial neural networks
MEHMET ALİ KÖLMEK
Yüksek Lisans
Türkçe
2012
Elektrik ve Elektronik MühendisliğiAnkara ÜniversitesiElektrik-Elektronik Mühendisliği Ana Bilim Dalı
YRD. DOÇ. DR. İSA NAVRUZ
- Elektrı̇k pı̇yasalarında elektrı̇k yük talebı̇ ve gün öncesı̇ elektrı̇k fı̇yat tahmı̇nı̇: Türkı̇ye uygulaması
Electricity load demand and day-ahead electricity price forecast in electricity markets: Implementation on Turkey
FAHRETTİN FİLİZ
- Elektrik piyasasında fiyatlandırma ve Türkiye elektrik piyasasında fiyat tahmini
Pricing in electricity market and price forecasting in Turkish electricity market
İBRAHİM MURAT BİCİL